library(BayesComposition)
library(ggplot2)
N <- 500
d <- 4
Gaussian Simulation
dat <- sim_compositional_data(N=N, d=d, likelihood = "gaussian",
additive_correlation = TRUE)
simPlotData <- data.frame(species=as.factor(rep(1:d, each=N)),
count=c(dat$y),
depth=rep(dat$X, times=d), alpha=c(dat$alpha))
gsim1 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25)
gsim2 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth by species") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25) +
facet_wrap( ~ species, ncol = 4)
multiplot(gsim1, gsim2, cols=1)
multi-logit simulation using basis functions without overdispersion
dat <- sim_compositional_data(N=N, d=d, likelihood = "multi-logit",
additive_correlation = FALSE,
multiplicative_correlation = TRUE)
y_dens <- matrix(0, N, d)
p_alpha <- matrix(0, N, d)
for (i in 1:N) {
y_dens[i, ] <- dat$y[i, ] / sum(dat$y[i, ])
p_alpha[i, ] <- exp(dat$mu + dat$zeta[i, ]) / sum(exp(dat$mu + dat$zeta[i, ]))
}
simPlotData <- data.frame(species=as.factor(rep(1:d, each=N)),
count=c(y_dens),
depth=rep(dat$X, times=d), alpha=c(p_alpha))
gsim1 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25)
gsim2 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth by species") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25) +
facet_wrap( ~ species, ncol = 4)
multiplot(gsim1, gsim2, cols=1)
multi-logit simulation using basis functions with overdispersion
dat <- sim_compositional_data(N=N, d=d, likelihood = "multi-logit",
additive_correlation = TRUE,
multiplicative_correlation = TRUE)
y_dens <- matrix(0, N, d)
p_alpha <- matrix(0, N, d)
for (i in 1:N) {
y_dens[i, ] <- dat$y[i, ] / sum(dat$y[i, ])
p_alpha[i, ] <- exp(dat$mu + dat$zeta[i, ]) / sum(exp(dat$mu + dat$zeta[i, ]))
}
simPlotData <- data.frame(species=as.factor(rep(1:d, each=N)),
count=c(y_dens),
depth=rep(dat$X, times=d), alpha=c(p_alpha))
gsim1 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25)
gsim2 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth by species") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25) +
facet_wrap( ~ species, ncol = 4)
multiplot(gsim1, gsim2, cols=1)
multi-logit simulation using Gaussian process without overdispersion
dat <- sim_compositional_data(N=N, d=d, likelihood = "multi-logit",
function_type = "gaussian-process",
correlation_function = "gaussian",
additive_correlation = FALSE,
multiplicative_correlation = TRUE)
y_dens <- matrix(0, N, d)
p_alpha <- matrix(0, N, d)
for (i in 1:N) {
y_dens[i, ] <- dat$y[i, ] / sum(dat$y[i, ])
p_alpha[i, ] <- exp(dat$mu + dat$zeta[i, ]) / sum(exp(dat$mu + dat$zeta[i, ]))
}
simPlotData <- data.frame(species=as.factor(rep(1:d, each=N)),
count=c(y_dens),
depth=rep(dat$X, times=d), alpha=c(p_alpha))
gsim1 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25)
gsim2 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth by species") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25) +
facet_wrap( ~ species, ncol = 4)
multiplot(gsim1, gsim2, cols=1)
multi-logit simulation using Gaussian process without overdispersion
dat <- sim_compositional_data(N=N, d=d, likelihood = "multi-logit",
function_type = "gaussian-process",
correlation_function = "gaussian",
additive_correlation = TRUE,
multiplicative_correlation = TRUE)
y_dens <- matrix(0, N, d)
p_alpha <- matrix(0, N, d)
for (i in 1:N) {
y_dens[i, ] <- dat$y[i, ] / sum(dat$y[i, ])
p_alpha[i, ] <- exp(dat$mu + dat$zeta[i, ]) / sum(exp(dat$mu + dat$zeta[i, ]))
}
simPlotData <- data.frame(species=as.factor(rep(1:d, each=N)),
count=c(y_dens),
depth=rep(dat$X, times=d), alpha=c(p_alpha))
gsim1 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25)
gsim2 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth by species") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25) +
facet_wrap( ~ species, ncol = 4)
multiplot(gsim1, gsim2, cols=1)
dirichlet-multinomial simulation using Gaussian process without overdispersion
dat <- sim_compositional_data(N=N, d=d, likelihood = "dirichlet-multinomial",
function_type = "gaussian-process",
correlation_function = "gaussian",
additive_correlation = FALSE,
multiplicative_correlation = TRUE)
y_dens <- matrix(0, N, d)
p_alpha <- matrix(0, N, d)
for (i in 1:N) {
y_dens[i, ] <- dat$y[i, ] / sum(dat$y[i, ])
p_alpha[i, ] <- exp(dat$mu + dat$zeta[i, ]) / sum(exp(dat$mu + dat$zeta[i, ]))
}
simPlotData <- data.frame(species=as.factor(rep(1:d, each=N)),
count=c(y_dens),
depth=rep(dat$X, times=d), alpha=c(p_alpha))
gsim1 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25)
gsim2 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth by species") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25) +
facet_wrap( ~ species, ncol = 4)
multiplot(gsim1, gsim2, cols=1)
dirichlet-multinomial simulation using Gaussian process without overdispersion
dat <- sim_compositional_data(N=N, d=d, likelihood = "dirichlet-multinomial",
function_type = "gaussian-process",
correlation_function = "gaussian",
additive_correlation = FALSE,
multiplicative_correlation = TRUE)
y_dens <- matrix(0, N, d)
p_alpha <- matrix(0, N, d)
for (i in 1:N) {
y_dens[i, ] <- dat$y[i, ] / sum(dat$y[i, ])
p_alpha[i, ] <- exp(dat$mu + dat$zeta[i, ]) / sum(exp(dat$mu + dat$zeta[i, ]))
}
simPlotData <- data.frame(species=as.factor(rep(1:d, each=N)),
count=c(y_dens),
depth=rep(dat$X, times=d), alpha=c(p_alpha))
gsim1 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25)
gsim2 <- ggplot(simPlotData, aes(x=depth, y=count, color=species, group=species)) +
geom_point(alpha=0.25) + theme(legend.position="none") +
ggtitle("Simulated response vs. depth by species") +
geom_line(aes(x=depth, y=alpha, col = species), simPlotData, lwd=1.25) +
facet_wrap( ~ species, ncol = 4)
multiplot(gsim1, gsim2, cols=1)
